Scholarlink

This research analyzes various machine learning classifiers for detecting security threats in wireless sensor networks (WSNs) using the WSN-DS dataset. It emphasizes performance metrics such as ROC AUC, accuracy, F1 score, precision, and recall. The proposed method addresses class imbalance effectively without resampling, achieving an accuracy of 99.69% with high precision and recall across all attack categories. To improve model transparency, SHAP and LIME were used for feature interpretability. Key features like SCH_S, Rank, and who_CH were identified as crucial for distinguishing between normal and attack instances. SHAP summary plots illustrate feature interactions, while LIME visualizations explain how these features impact individual classifications, especially in flooding attacks. The results highlight the model’s robustness and stress the importance of explainability in machine learning-based WSN attack detection, offering a valuable approach for improving the reliability and trustworthiness of detection systems in critical applications.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Explainability in Security Attack Detection for Wireless Sensor Networks with Imbalanced Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Deepa Krishnan

摘要

This research analyzes various machine learning classifiers for detecting security threats in wireless sensor networks (WSNs) using the WSN-DS dataset. It emphasizes performance metrics such as ROC AUC, accuracy, F1 score, precision, and recall. The proposed method addresses class imbalance effectively without resampling, achieving an accuracy of 99.69% with high precision and recall across all attack categories. To improve model transparency, SHAP and LIME were used for feature interpretability. Key features like SCH_S, Rank, and who_CH were identified as crucial for distinguishing between normal and attack instances. SHAP summary plots illustrate feature interactions, while LIME visualizations explain how these features impact individual classifications, especially in flooding attacks. The results highlight the model’s robustness and stress the importance of explainability in machine learning-based WSN attack detection, offering a valuable approach for improving the reliability and trustworthiness of detection systems in critical applications.